{"id":"https://openalex.org/W4414836780","doi":"https://doi.org/10.52202/085713-4461","title":"KVLink: Accelerating Large Language Models via Efficient KV Cache Reuse","display_name":"KVLink: Accelerating Large Language Models via Efficient KV Cache Reuse","publication_year":2025,"publication_date":"2025-01-01","ids":{"openalex":"https://openalex.org/W4414836780","doi":"https://doi.org/10.52202/085713-4461"},"language":"en","primary_location":{"id":"doi:10.52202/085713-4461","doi":"https://doi.org/10.52202/085713-4461","is_oa":false,"landing_page_url":"https://doi.org/10.52202/085713-4461","pdf_url":null,"source":{"id":"https://openalex.org/S4363606243","display_name":"neural information processing systems","issn_l":"1049-5258","issn":["1049-5258"],"is_oa":false,"is_in_doaj":false,"is_core":true,"listed_in":["cwts-core","norway-2"],"host_organization":null,"host_organization_name":null,"host_organization_lineage":[],"host_organization_lineage_names":[],"type":"conference"},"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Advances in Neural Information Processing Systems 38","raw_type":"proceedings-article"},"volume":null,"type":"conference-paper","indexed_in":["arxiv","crossref","datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://arxiv.org/pdf/2502.16002","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5102024457","display_name":"Jingbo Yang","orcid":"https://orcid.org/0000-0002-6068-1325","observed_orcids":["https://orcid.org/0000-0002-6068-1325"]},"institutions":[{"id":"https://openalex.org/I154570441","display_name":"University of California, Santa Barbara","ror":"https://ror.org/02t274463","country_code":"US","type":"education","lineage":["https://openalex.org/I154570441"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Jingbo Yang","raw_affiliation_strings":["University of California, Santa Barbara"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"University of California, Santa Barbara","institution_ids":["https://openalex.org/I154570441"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5111136892","display_name":"Bairu Hou","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I154570441","display_name":"University of California, Santa Barbara","ror":"https://ror.org/02t274463","country_code":"US","type":"education","lineage":["https://openalex.org/I154570441"]},{"id":"https://openalex.org/I4210153776","display_name":"Apple (United States)","ror":"https://ror.org/059hsda18","country_code":"US","type":"company","lineage":["https://openalex.org/I4210153776"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Bairu Hou","raw_affiliation_strings":["UCSB / Apple"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"UCSB / Apple","institution_ids":["https://openalex.org/I154570441","https://openalex.org/I4210153776"]}]},{"author_position":"middle","author":{"id":null,"display_name":"Wei Wei","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I181877577","display_name":"Shanxi University","ror":"https://ror.org/03y3e3s17","country_code":"CN","type":"education","lineage":["https://openalex.org/I181877577"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Wei Wei","raw_affiliation_strings":["shanxi university"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"shanxi university","institution_ids":["https://openalex.org/I181877577"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5027122418","display_name":"Yujia Bao","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I1310439424","display_name":"Accenture (Switzerland)","ror":"https://ror.org/041r3e346","country_code":"CH","type":"company","lineage":["https://openalex.org/I1310439424","https://openalex.org/I4210093804"]},{"id":"https://openalex.org/I4210099672","display_name":"Accenture (United States)","ror":"https://ror.org/013g16z83","country_code":"US","type":"company","lineage":["https://openalex.org/I4210093804","https://openalex.org/I4210099672"]}],"countries":["CH","US"],"is_corresponding":false,"raw_author_name":"Yujia Bao","raw_affiliation_strings":["Accenture"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Accenture","institution_ids":["https://openalex.org/I1310439424","https://openalex.org/I4210099672"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5112248869","display_name":"Shiyu Chang","orcid":"https://orcid.org/0009-0006-2943-375X","observed_orcids":["https://orcid.org/0009-0006-2943-375X"]},"institutions":[{"id":"https://openalex.org/I154570441","display_name":"University of California, Santa Barbara","ror":"https://ror.org/02t274463","country_code":"US","type":"education","lineage":["https://openalex.org/I154570441"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Shiyu Chang","raw_affiliation_strings":["UC Santa Barbara"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"UC Santa Barbara","institution_ids":["https://openalex.org/I154570441"]}]}],"institutions":[],"countries_distinct_count":3,"institutions_distinct_count":5,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":4.6703,"has_fulltext":false,"cited_by_count":5,"citation_normalized_percentile":{"value":0.95608516,"is_in_top_1_percent":false,"is_in_top_10_percent":true},"cited_by_percentile_year":{"min":98,"max":99},"biblio":{"volume":null,"issue":null,"first_page":"148312","last_page":"148339"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10028","display_name":"Topic Modeling","score":0.899399995803833,"subfield":{"id":"https://openalex.org/subfields/1702","display_name":"Artificial Intelligence"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},"topics":[{"id":"https://openalex.org/T10028","display_name":"Topic Modeling","score":0.899399995803833,"subfield":{"id":"https://openalex.org/subfields/1702","display_name":"Artificial Intelligence"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},{"id":"https://openalex.org/T10181","display_name":"Natural Language Processing Techniques","score":0.06669999659061432,"subfield":{"id":"https://openalex.org/subfields/1702","display_name":"Artificial Intelligence"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},{"id":"https://openalex.org/T10286","display_name":"Information Retrieval and Search Behavior","score":0.01769999973475933,"subfield":{"id":"https://openalex.org/subfields/1710","display_name":"Information Systems"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}}],"keywords":[{"id":"https://openalex.org/keywords/large-language-models","display_name":"large language models","score":0.9649999737739563},{"id":"https://openalex.org/keywords/kv-cache-compression","display_name":"KV cache compression","score":0.5929999947547913},{"id":"https://openalex.org/keywords/question-answering","display_name":"question answering","score":0.6679999828338623},{"id":"https://openalex.org/keywords/time-to-first-token","display_name":"time-to-first-token","score":0.8370000123977661},{"id":"https://openalex.org/keywords/document-retrieval","display_name":"document retrieval","score":0.5389999747276306},{"id":"https://openalex.org/keywords/inference-efficiency","display_name":"inference efficiency","score":0.45399999618530273}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.885200023651123},{"id":"https://openalex.org/C115537543","wikidata":"https://www.wikidata.org/wiki/Q165596","display_name":"Cache","level":2,"score":0.8478000164031982},{"id":"https://openalex.org/C206588197","wikidata":"https://www.wikidata.org/wiki/Q846574","display_name":"Reuse","level":2,"score":0.7123000025749207},{"id":"https://openalex.org/C2779343474","wikidata":"https://www.wikidata.org/wiki/Q3109175","display_name":"Context (archaeology)","level":2,"score":0.5723999738693237},{"id":"https://openalex.org/C48044578","wikidata":"https://www.wikidata.org/wiki/Q727490","display_name":"Scalability","level":2,"score":0.5719000101089478},{"id":"https://openalex.org/C25536678","wikidata":"https://www.wikidata.org/wiki/Q5015977","display_name":"Cache invalidation","level":5,"score":0.5450000166893005},{"id":"https://openalex.org/C38556500","wikidata":"https://www.wikidata.org/wiki/Q13404475","display_name":"Cache algorithms","level":4,"score":0.4948999881744385},{"id":"https://openalex.org/C2779960059","wikidata":"https://www.wikidata.org/wiki/Q7113681","display_name":"Overhead (engineering)","level":2,"score":0.47940000891685486},{"id":"https://openalex.org/C2776214188","wikidata":"https://www.wikidata.org/wiki/Q408386","display_name":"Inference","level":2,"score":0.4745999872684479},{"id":"https://openalex.org/C173608175","wikidata":"https://www.wikidata.org/wiki/Q232661","display_name":"Parallel computing","level":1,"score":0.4731000065803528},{"id":"https://openalex.org/C26517878","wikidata":"https://www.wikidata.org/wiki/Q228039","display_name":"Key (lock)","level":2,"score":0.4072999954223633},{"id":"https://openalex.org/C66746571","wikidata":"https://www.wikidata.org/wiki/Q1134833","display_name":"ENCODE","level":3,"score":0.4025000035762787},{"id":"https://openalex.org/C167713795","wikidata":"https://www.wikidata.org/wiki/Q352090","display_name":"Smart Cache","level":5,"score":0.3837999999523163},{"id":"https://openalex.org/C113166858","wikidata":"https://www.wikidata.org/wiki/Q5015981","display_name":"Cache pollution","level":5,"score":0.3601999878883362},{"id":"https://openalex.org/C189783530","wikidata":"https://www.wikidata.org/wiki/Q352090","display_name":"CPU cache","level":3,"score":0.3479999899864197},{"id":"https://openalex.org/C51185590","wikidata":"https://www.wikidata.org/wiki/Q1017228","display_name":"Bus sniffing","level":5,"score":0.3456999957561493},{"id":"https://openalex.org/C137293760","wikidata":"https://www.wikidata.org/wiki/Q3621696","display_name":"Language model","level":2,"score":0.3400000035762787},{"id":"https://openalex.org/C125411270","wikidata":"https://www.wikidata.org/wiki/Q18653","display_name":"Encoding (memory)","level":2,"score":0.31520000100135803},{"id":"https://openalex.org/C201148951","wikidata":"https://www.wikidata.org/wiki/Q5015976","display_name":"Cache coloring","level":4,"score":0.29809999465942383},{"id":"https://openalex.org/C57273362","wikidata":"https://www.wikidata.org/wiki/Q576722","display_name":"Decoding methods","level":2,"score":0.2833000123500824}],"mesh":[],"locations_count":3,"locations":[{"id":"doi:10.52202/085713-4461","doi":"https://doi.org/10.52202/085713-4461","is_oa":false,"landing_page_url":"https://doi.org/10.52202/085713-4461","pdf_url":null,"source":{"id":"https://openalex.org/S4363606243","display_name":"neural information processing systems","issn_l":"1049-5258","issn":["1049-5258"],"is_oa":false,"is_in_doaj":false,"is_core":true,"listed_in":["cwts-core","norway-2"],"host_organization":null,"host_organization_name":null,"host_organization_lineage":[],"host_organization_lineage_names":[],"type":"conference"},"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Advances in Neural Information Processing Systems 38","raw_type":"proceedings-article"},{"id":"pmh:oai:arXiv.org:2502.16002","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2502.16002","pdf_url":"https://arxiv.org/pdf/2502.16002","source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":null,"license_id":null,"version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},{"id":"doi:10.48550/arxiv.2502.16002","doi":"https://doi.org/10.48550/arxiv.2502.16002","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2502.16002","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Preprint"}],"best_oa_location":{"id":"pmh:oai:arXiv.org:2502.16002","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2502.16002","pdf_url":"https://arxiv.org/pdf/2502.16002","source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":null,"license_id":null,"version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},"sustainable_development_goals":[],"study_designs":[],"awards":[],"funders":[],"has_content":{"grobid_xml":false,"pdf":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":["https://openalex.org/W7154612387","https://openalex.org/W4417287129","https://openalex.org/W6979369520","https://openalex.org/W7106683035","https://openalex.org/W7170093618","https://openalex.org/W4414197699","https://openalex.org/W7117152659","https://openalex.org/W7134860038","https://openalex.org/W4417092586","https://openalex.org/W7172534266"],"abstract_inverted_index":{"We":[0],"describe":[1],"KVLink,":[2],"an":[3,155],"approach":[4,169],"for":[5,46,107,188],"efficient":[6,186],"key-value":[7],"(KV)":[8],"cache":[9,68,121,198,203],"reuse":[10,89],"in":[11,33],"large":[12],"language":[13],"models":[14],"(LLMs).":[15],"In":[16,53],"many":[17],"LLM":[18,179],"applications,":[19],"different":[20],"inputs":[21],"can":[22,193],"share":[23],"overlapping":[24],"context,":[25],"such":[26,63],"as":[27],"the":[28,37,43,66,77,86,98,119,126,210],"same":[29],"retrieved":[30,81],"document":[31,71],"appearing":[32],"multiple":[34],"queries.":[35],"However,":[36],"LLMs":[38],"still":[39],"need":[40],"to":[41,50,61,88,124,136,174,177,200],"encode":[42],"entire":[44],"context":[45,189],"each":[47,70,108],"query,":[48],"leading":[49],"redundant":[51],"computation.":[52],"this":[54],"paper,":[55],"we":[56],"investigate":[57],"a":[58,183],"new":[59],"strategy":[60],"eliminate":[62],"inefficiency,":[64],"where":[65],"KV":[67,78,103,120,166,197],"of":[69,80,93,118,157],"is":[72],"precomputed":[73,165],"independently.":[74],"During":[75],"inference,":[76,180],"caches":[79,104],"documents":[82],"are":[83],"concatenated,":[84],"allowing":[85],"model":[87],"cached":[90],"representations":[91],"instead":[92],"recomputing":[94],"them.":[95],"To":[96],"mitigate":[97],"performance":[99],"degradation":[100],"when":[101],"using":[102,132],"computed":[105],"independently":[106,140],"document,":[109],"KVLink":[110,149,192],"introduces":[111],"two":[112],"key":[113],"techniques:":[114],"adjusting":[115],"positional":[116],"embeddings":[117],"at":[122],"inference":[123],"match":[125],"global":[127],"position":[128],"after":[129],"concatenation,":[130],"and":[131,185,205],"trainable":[133],"special":[134],"tokens":[135],"restore":[137],"self-attention":[138],"across":[139,144],"encoded":[141],"documents.":[142],"Experiments":[143],"7":[145],"datasets":[146],"demonstrate":[147],"that":[148],"improves":[150],"question":[151],"answering":[152],"accuracy":[153],"by":[154,163,172],"average":[156],"4%":[158],"over":[159],"state-of-the-art":[160],"methods.":[161],"Furthermore,":[162],"leveraging":[164],"caches,":[167],"our":[168],"reduces":[170],"time-to-first-token":[171],"up":[173],"96%":[175],"compared":[176],"standard":[178],"making":[181],"it":[182],"scalable":[184],"solution":[187],"reuse.":[190],"Additionally,":[191],"be":[194],"combined":[195],"with":[196],"compression":[199],"further":[201],"save":[202],"loading":[204],"storage":[206],"overhead":[207],"while":[208],"outperforming":[209],"baselines.":[211]},"counts_by_year":[{"year":2026,"cited_by_count":5}],"updated_date":"2026-11-03T10:50:19.383937","created_date":"2025-10-10T00:00:00"}